Method and system for real-time regulation and control of postures of tunnel lining bench vehicle based on multi-sensor data

Abstract

The invention relates to the technical field of tunnel construction mechanical automation, and discloses a tunnel lining bench car posture real-time regulation and control method and system based on multi-sensing data. The method comprises the following steps: collecting multi-dimensional pose data in a trolley movement process to construct a space-time associated sequence; performing multi-scale segmentation on the sequence, identifying a feature data segment, analyzing a change mode and an extreme point evolution law of the feature data segment, and generating a periodic regularity coefficient; constructing a spatial-temporal characteristic matrix of the pose data, extracting principal components through matrix decomposition, calculating concentration degree and dispersity indexes, and fusing internal correlation characteristics and a periodic regularity coefficient to obtain a quality evaluation coefficient; and dynamically adjusting a pose data processing strategy based on the quality evaluation coefficient to realize accurate pose regulation and control of the trolley. The method can deeply excavate the internal law of the pose data, is adaptive to the data quality change, and effectively improves the precision and stability of platform and vehicle pose regulation and control.

Description

Technical Field

[0001] This application relates to the field of automation technology for tunnel construction machinery, specifically a method and system for real-time control of the position and posture of a tunnel lining trolley based on multi-sensor data. Background Technology

[0002] During tunnel lining construction, the positioning accuracy of the trolley directly affects the lining quality and structural safety. Existing technologies typically rely on multiple sensors mounted on the trolley to collect positioning data, and then use pre-programmed logic or classic PID control algorithms to adjust the trolley's support cylinders and other components to position it within the designed contour. These conventional methods treat the collected positioning data as a uniform time series, performing mean filtering and deviation calculations for the entire data or a fixed time window, and triggering control commands based on simple threshold judgments.

[0003] This conventional approach has its limitations. Due to the complex environment within tunnels, the motion characteristics of the trolley vary significantly across different stages, such as travel, positioning, and pouring, resulting in pose data exhibiting strong non-stationarity and multi-scale characteristics. Fixed-window data processing struggles to accurately capture the dynamic patterns of the trolley during specific motion stages, leading to system responses that either lag behind actual pose changes or become misadjusted or oscillate due to excessive sensitivity to noise. Existing technologies lack quantitative analysis of the deep-seated patterns within pose data sequences, failing to distinguish between accidental fluctuations and inherent motion patterns. This results in simplistic and rigid control strategies that are ill-suited to complex construction conditions.

[0004] Existing methods lack the ability to online evaluate the quality of pose data itself. Control systems typically assume all input data has equal reliability and process it using fixed algorithm parameters. In practical applications, sensor readings are affected by factors such as mechanical vibration, environmental interference, and momentary occlusion, causing dynamic changes in data quality. When data anomalies occur or the signal-to-noise ratio decreases, rigid processing strategies cannot adaptively adjust, potentially leading to erroneous decisions based on low-quality data. This not only affects control accuracy but may even cause equipment failure. How to dynamically evaluate the quality of pose data and optimize processing strategies accordingly is a key challenge in achieving high-precision, high-stability real-time control. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for real-time control of the position and orientation of a tunnel lining trolley based on multi-sensor data, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this application provides a method for real-time pose control of a tunnel lining trolley based on multi-sensor data. The method includes: collecting multi-dimensional pose measurement data of the tunnel lining trolley during its movement to construct a spatiotemporally correlated pose data sequence; performing multi-scale segmentation on the pose data sequence to identify a set of feature data segments reflecting the trolley's motion patterns; analyzing the change patterns of each data segment in the feature data segment set, calculating the trend consistency index and fluctuation coordination index within each data segment; detecting the distribution pattern of extreme points in the feature data segment set, extracting the gradient and time interval features between extreme points to form an extreme value evolution pattern description; and comprehensively analyzing the feature data... Based on the overall distribution characteristics of the segment set, the extreme value evolution pattern description, and the change pattern of each data segment, the periodic regularity coefficient of the pose data sequence is generated based on the trend consistency index and the fluctuation coordination index. A spatiotemporal feature matrix of the pose data sequence is constructed, and the main feature components of the sequence are extracted using matrix decomposition technology. The concentration and dispersion indices of the feature components are calculated. The internal correlation characteristics of the spatiotemporal feature matrix, the periodic regularity coefficient of the pose data sequence, and the concentration and dispersion indices of the feature components are integrated to obtain the quality assessment coefficient of the pose data sequence. Based on the quality assessment coefficient, the pose data processing strategy is dynamically adjusted to achieve pose control of the tunnel lining trolley.

[0007] Preferably, the multi-scale segmentation processing of the pose data sequence to identify a set of feature data segments reflecting the motion law of the trolley includes: scanning the pose data sequence using a sliding window detection algorithm to identify all extreme points in the sequence, including local peak points and valley points; segmenting the pose data sequence into multiple continuous data segments using adjacent valley points as boundaries; calculating the data change amplitude of each data segment; grouping the change amplitudes of all data segments using a density clustering algorithm; and selecting the group of data segments with the largest change amplitude as the set of feature data segments.

[0008] Preferably, the calculation method for the trend consistency index and the fluctuation coordination index includes: for each data segment in the set of feature data segments, dividing the data segment into forward sub-segments and backward sub-segments with the maximum value of the data segment as the dividing point; calculating the monotonicity of the data change trend of the forward sub-segment and the monotonicity of the data change trend of the backward sub-segment, and using the absolute value of the difference between the two monotonicity levels as the trend consistency index; analyzing the periodic characteristics of the data fluctuation of the forward sub-segment and the periodic characteristics of the data fluctuation of the backward sub-segment, calculating the matching degree of the fluctuation cycles of the two sub-segments, and using the matching degree of the fluctuation cycles of the two sub-segments as the fluctuation coordination index.

[0009] Preferably, the method for calculating the degree of monotonicity includes: performing differential processing on the sub-segment data, counting the number and length of consecutive segments with the same sign in the differential sequence; and calculating the overall monotonicity score of the sub-segment data based on the distribution characteristics of the consecutive segments with the same sign to obtain the degree of monotonicity.

[0010] Preferably, the method for extracting the extreme value evolution pattern description includes: arranging all extreme points in the feature data segment set in chronological order to construct an extreme point sequence; calculating the numerical change gradient sequence of adjacent extreme points in the extreme point sequence and analyzing the stability characteristics of the change gradient sequence; statistically analyzing the distribution pattern of time intervals between extreme points and calculating the coefficient of variation of the time intervals; and combining the stability characteristics of the change gradient sequence and the coefficient of variation of the time intervals to form an extreme value evolution pattern description.

[0011] Preferably, the method for generating the periodicity coefficient includes: calculating the weighted average of the trend consistency index and fluctuation coordination index of all data segments in the feature data segment set; analyzing the morphological similarity between data segments in the feature data segment set and calculating the average similarity score of the feature data segment set; normalizing the stability characteristics and coefficient of variation in the extreme value evolution pattern description to obtain the extreme value change regularity; and merging the weighted average, average similarity score, and extreme value change regularity to generate the periodicity coefficient.

[0012] Preferably, the method for calculating the concentration index and dispersion index of the feature components includes: performing singular value decomposition on the spatiotemporal feature matrix, arranging the singular values ​​in descending order to form a singular value spectrum; using an adaptive threshold algorithm to divide the singular value spectrum into primary component regions and secondary component regions; calculating the cumulative contribution rate of singular values ​​in the primary component regions, and using the cumulative contribution rate of singular values ​​in the primary component regions as the concentration index; analyzing the degree of difference in the distribution of singular values ​​between the primary component regions and the secondary component regions, and using the degree of difference as the dispersion index.

[0013] Preferably, the method for calculating the quality assessment coefficient includes: calculating the correlation matrix between the row vectors of the spatiotemporal feature matrix and analyzing the internal structural features of the matrix; coupling the internal structural features with the periodicity coefficient to obtain the regularity score of the sequence; combining and transforming the concentration index and dispersion index of the feature components to obtain the feature integrity score of the sequence; and combining the regularity score and the feature integrity score to generate the quality assessment coefficient through weighted fusion.

[0014] Preferably, the method for dynamically adjusting the pose data processing strategy based on the quality assessment coefficient includes: determining the intensity parameter of the pose data filtering according to the quality assessment coefficient, and establishing a mapping relationship between the filtering intensity and the quality assessment coefficient; calculating the real-time pose state of the trolley based on the filtered pose data; generating a pose adjustment command based on the deviation between the real-time pose state and the desired pose; and executing the pose adjustment command through a closed-loop control system to realize the pose control of the tunnel lining trolley.

[0015] Preferably, this application also includes a real-time position and orientation control system for a tunnel lining trolley based on multi-sensor data. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the real-time position and orientation control method for a tunnel lining trolley based on multi-sensor data as described above.

[0016] Compared with the prior art, the beneficial effects of this application are as follows: By performing multi-scale segmentation on the pose data sequence, the system identifies sets of characteristic data segments corresponding to different motion stages of the trolley. It then analyzes the consistency of trends and the coordination of fluctuations within each data segment, extracts the gradient and time interval patterns between extreme points, and finally generates periodicity coefficients that quantitatively describe the periodicity and stability of the trolley's motion. This method can deeply analyze the dynamic evolution patterns of pose data at both macroscopic and microscopic levels, accurately capture the motion patterns of the trolley under specific working conditions, and enable the control system to go beyond simple responses to instantaneous deviations. It allows for predictive perception of the trolley's motion trends and periodic patterns, providing richer and more accurate data for control strategies and enhancing the ability to characterize complex nonlinear motion behaviors.

[0017] A spatiotemporal feature matrix of pose data is constructed. Its main feature components are extracted through matrix decomposition, and its concentration and dispersion indices are calculated. These indices are then integrated with the internal correlation characteristics and periodicity coefficients of the sequence to generate a comprehensive quality assessment coefficient. This coefficient, as a meta-level parameter, reflects the reliability and characteristic patterns of the current pose data in real time. Based on this coefficient, the pose data processing strategy is dynamically adjusted, enabling the system to employ a more sensitive strategy to pursue accuracy when data quality is high, and to switch to a more robust strategy to ensure safety when data quality is compromised. This mechanism endows the control system with adaptive capabilities to dynamic changes in data quality, effectively avoiding erroneous adjustments based on low-quality or abnormal data, and improving the stability and reliability of the entire system under real-world, harsh operating conditions. Attached Figure Description

[0018] Figure 1 A schematic diagram illustrating the working principle of the real-time position and orientation control method for tunnel lining trolley based on multi-sensor data provided in this application embodiment; Figure 2 A flowchart of multi-scale segmentation processing and feature data segment recognition provided in the embodiments of this application; Figure 3 A flowchart illustrating the calculation of the trend consistency index and volatility coordination index provided in the embodiments of this application; Figure 4 A comparative analysis chart of trend consistency index and fluctuation coordination index of tunnel lining trolley characteristic data segments provided in the embodiments of this application; Figure 5 A graph showing the periodic regularity coefficient variation of the tunnel lining trolley position and posture data provided in this application embodiment. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 The present invention provides a method for real-time control of the position and posture of a tunnel lining trolley based on multi-sensor data, and the specific implementation method is as follows.

[0021] This method collects multi-dimensional pose measurement data of a tunnel lining trolley during its movement, constructs a spatiotemporally correlated pose data sequence, and then processes and analyzes the sequence to achieve precise control of the trolley's pose. The construction of the pose data sequence involves multiple sensors, such as inertial measurement units, laser rangefinders, and vision sensors, which simultaneously collect data and perform timestamp alignment and spatial coordinate transformation to form a sequence containing position, attitude, and motion parameters. Subsequently, the sequence undergoes multi-scale segmentation to identify a set of characteristic data segments reflecting the trolley's motion patterns. This step is achieved through extreme point detection and cluster analysis. Next, the variation patterns of each data segment in the characteristic data segment set are analyzed, and trend consistency indices and fluctuation coordination indices are calculated. These indices are based on the monotonicity and periodicity of the data segments. Simultaneously, the distribution patterns of extreme points in the characteristic data segment set are detected, and variation gradient and time interval features are extracted to form an extreme value evolution pattern description. Based on the above analysis, the overall distribution of the characteristic data segment set, the extreme value evolution pattern, and the data segment variation pattern are combined to generate a periodicity coefficient for the pose data sequence based on the trend consistency index and fluctuation coordination index. Furthermore, a spatiotemporal feature matrix of the pose data sequence is constructed. Key feature components are extracted using matrix factorization, and concentration and dispersion indices of these components are calculated. Finally, the internal correlation of the spatiotemporal feature matrix, the periodicity coefficient of the pose data sequence, and the concentration and dispersion indices of the feature components are fused to obtain a quality assessment coefficient for the pose data sequence. This coefficient is used to dynamically adjust pose data processing strategies, such as filtering parameters and control commands, thereby achieving real-time pose control of the tunnel lining trolley. The entire process is implemented using an embedded system or industrial computer, ensuring low latency and high reliability.

[0022] Example 1: See Figure 2In practical implementation, the real-time pose control method for tunnel lining trolleys based on multi-sensor data involves multi-scale segmentation of the pose data sequence to identify a set of feature data segments. This process begins by scanning the pose data sequence using a sliding window detection algorithm. The sliding window detection algorithm sets the window size according to the motion characteristics of the tunnel lining trolley and the data sampling frequency. The window size is usually matched with the typical motion cycle of the trolley to ensure complete coverage of the fluctuation pattern. During the operation of the sliding window detection algorithm, the algorithm compares the data value within the window with the values ​​of adjacent points point by point, identifying all local peak points and valley points. Local peak points are points where the data value is greater than the adjacent points before and after, and local valley points are points where the data value is less than the adjacent points before and after. Threshold filtering is used during identification to eliminate false extreme points caused by sensor noise. The threshold is dynamically adjusted based on historical data or real-time statistics. In some embodiments, the sliding window detection algorithm can be combined with multi-resolution analysis to adapt to pose changes at different time scales by adjusting the window size, thereby improving the robustness of extreme point detection. In practice, after identifying extreme points, the pose data sequence is divided into multiple continuous data segments using adjacent valley points as boundaries. Each data segment represents the pose change trajectory of the tunnel lining trolley within a complete motion cycle. The segmentation process ensures that the start and end points of the data segments are both valley points, thus forming coherent periodic units. The segmentation of data segments is automatically performed based on the temporal position of extreme points. The boundaries of each segment are marked by an index and stored as independent data blocks for subsequent processing.

[0023] In practical implementation, calculating the amplitude of data change for each data segment is a crucial step in the segmentation process. The amplitude is derived from the difference between the maximum and minimum values ​​within the data segment. These maximum and minimum values ​​are extracted from the data points within the segment. The amplitude value quantifies the intensity of the pose fluctuation of the tunnel lining trolley during that motion cycle. After amplitude calculation, a density clustering algorithm is used to group the amplitude changes of all data segments. This algorithm employs a distance-based clustering method such as DBSCAN, automatically determining the number and boundaries of clusters based on the distribution density of amplitude values. DBSCAN requires setting neighborhood radius and minimum sample size parameters, which are configured based on the historical distribution or real-time statistics of amplitude values. The density clustering algorithm groups data segments with similar amplitude changes into the same group, forming multiple clusters. The clustering results reflect the heterogeneity of pose change patterns. In some embodiments, the density clustering algorithm can be replaced by hierarchical clustering or K-means clustering, but it must be ensured that the clustering process can adapt to the distribution characteristics of the amplitude values. In practice, the data segment set with the largest change amplitude is selected as the feature data segment set. The feature data segment set contains the motion cycle with the most significant pose changes. The selection criterion is based on the comparison of the average or maximum amplitude values ​​of each group after clustering. The group with the largest amplitude represents the key change stage in the movement of the tunnel lining trolley. The identification of the feature data segment set provides a foundation for subsequent analysis, ensuring a focus on high-dynamic pose data.

[0024] Optionally, the sliding window detection algorithm can integrate multi-sensor data fusion technology to simultaneously process data streams from the inertial measurement unit and laser rangefinder while scanning the pose data sequence, thereby enhancing the accuracy of extreme point detection. It is understood that the extreme point identification process needs to consider real-time interference from the tunnel environment, and adaptive filtering preprocesses the data sequence to reduce false detections. In specific implementation, data segmentation and amplitude calculation are performed by the real-time computing module of the embedded system. The module adopts a pipelined architecture to process continuous data streams, ensuring low latency. The grouping results of the density clustering algorithm are dynamically updated to adapt to changes in the movement mode of the tunnel lining trolley, and the clustering parameters are automatically adjusted according to new data. Optionally, the selection of the feature data segment set can be combined with temporal continuity constraints, retaining only temporally coherent data segment groups to improve reliability. It is understood that multi-scale segmentation processing as a whole relies on efficient memory management and parallel computing technology to process large-scale pose data sequences.

[0025] In practical implementation, the sliding window detection algorithm sets the window size based on the kinematic model of the tunnel lining trolley. The model predicts the typical cycle length of the trolley, and the window size is adjusted accordingly to a fixed or dynamic value. Dynamic adjustment is achieved by monitoring the autocorrelation characteristics of the data sequence. After extreme point identification, the data segmentation algorithm verifies the integrity of each segment, ensuring that each segment contains complete rising and falling phases; incomplete segments are discarded or merged. In practical implementation, the calculation of change amplitude includes not only numerical differences but can also introduce integral amplitude or root mean square amplitude as supplementary indicators to more comprehensively quantify the fluctuation intensity. The execution of the density clustering algorithm involves standardization of amplitude values ​​to eliminate the influence of dimensions. The clustering results are visualized for manual verification, but in automated systems, only the algorithm output is relied upon. The final determination of the feature data segment set is achieved through a voting mechanism or weighted scoring, combining amplitude values ​​and other features such as segment length to ensure that the selected set is highly representative.

[0026] In practical implementation, the overall process of multi-scale segmentation is embedded in the real-time control system. The system periodically executes the segmentation algorithm, processing the latest pose data sequence each time, with the sequence length set according to the control cycle. The computational complexity of the sliding window detection algorithm is reduced through optimization algorithms such as the two-pointer technique to meet real-time requirements. In the extreme point identification stage, a multinomial fitting is used to smooth the data sequence, reducing the impact of noise on extreme point detection. After data segmentation, each segment is assigned a unique identifier, and metadata such as start and end times and amplitude values ​​are stored. Parameter tuning of the density clustering algorithm is completed through offline or online learning. Offline learning uses historical data to train parameters, while online learning adjusts them according to the real-time data distribution. The selection results of the feature data segment set are directly passed to the subsequent analysis module for calculating trend consistency and fluctuation coordination indices.

[0027] In practical implementation, the sliding window detection algorithm can be extended to multi-dimensional detection, simultaneously processing multiple channels of pose data sequences, such as position and attitude data. Extreme point identification is based on channel fusion scores. Data segmentation considers multi-channel synchronization, ensuring segment boundaries are consistent across all channels. The amplitude calculation aggregates multi-channel amplitude values, employing a weighted average or maximum value strategy. Density clustering algorithms process multi-dimensional amplitude vectors, using Mahalanobis or Euclidean distance for clustering. The selection of the feature data segment set is based on multi-channel comprehensive scoring, fusing amplitude values ​​and inter-channel correlation. Understandably, the implementation of multi-scale segmentation requires efficient numerical computing libraries, such as BLAS or NumPy, to ensure processing speed.

[0028] In practical implementation, the robustness of the segmentation process is enhanced through an anomaly detection mechanism. Anomaly detection identifies invalid data segments, such as those caused by sensor malfunctions, and excludes these segments before clustering. The threshold setting of the sliding window detection algorithm is based on the principle of statistical process control, dynamically updating the threshold to cope with data drift. The results of extreme point identification are checked for temporal consistency to eliminate isolated extreme points. The data segmentation algorithm handles boundary cases, such as segments at the beginning and end of a sequence, ensuring integrity through padding or truncation. Normalization is introduced in the calculation of variation amplitude to make amplitude values ​​comparable across different data segments. The output of the density clustering algorithm includes clustering quality and confidence scores; low-confidence clustering triggers re-clustering. After selecting the feature data segment set, the segments within the set are sorted and indexed for easy subsequent rapid retrieval.

[0029] In practical implementation, performance monitoring of multi-scale segmentation processing is achieved through log recording and real-time metrics, including processing latency and accuracy, for system optimization. The sliding window detection algorithm employs incremental calculation using a sliding window to avoid repeatedly scanning the data sequence. Extreme point identification, combined with multi-scale analysis, detects extreme points at different time granularities to improve coverage. Data segmentation utilizes extreme point type information, such as alternating peak and valley patterns, to verify the segmentation logic. Variation amplitude calculation supports batch processing, calculating the amplitude values ​​of all segments at once. The density clustering algorithm automatically determines the number of clusters, using the elbow rule or silhouette coefficient for optimization. Persistent storage of the feature data segment set ensures continuity after system restart. The entire segmentation processing module is tightly integrated with the data acquisition and control module, forming a closed-loop control pipeline.

[0030] Example 2: See Figure 3In practical implementation, the changing patterns of each data segment in the set of characteristic data segments are analyzed, and trend consistency and fluctuation coordination indices are calculated. For each data segment in the set of characteristic data segments, the segment is divided into a forward sub-segment and a backward sub-segment, with the maximum value point as the dividing point. The maximum value point is determined by comparing the values ​​of all data points within the segment and is the point with the largest value in the segment. The dividing operation is achieved by locating the index position of the maximum value point. The forward sub-segment contains all data points from the start point of the data segment to the maximum value point, and the backward sub-segment contains all data points from the maximum value point to the end point of the data segment. The division ensures that the two sub-segments are continuous and cover the entire data segment. The trend consistency index is calculated based on the difference in the monotonicity of the data change trend in the forward sub-segment and the backward sub-segment. The monotonicity reflects the degree of overall increase or decrease in the sub-segment data. In practice, the calculation of monotonicity begins with differential processing of the sub-segment data. Differential processing calculates the numerical difference between adjacent data points in a sub-segment, forming a difference sequence. Positive values ​​in the difference sequence indicate an upward trend, while negative values ​​indicate a downward trend. During specific construction phases, the trolley maintains a relatively stable motion trend, such as uniform speed or smooth lifting and lowering. This corresponds to continuous upward or downward changes in the posture data, forming consecutive segments with the same sign in the difference sequence. Even if the trolley's motion experiences slight fluctuations, as long as the overall upward or downward trend remains unchanged, the sign of the difference sequence will not frequently flip, still forming corresponding consecutive segments with the same sign. However, minor noise from the sensors or slight interference from the tunnel environment may cause interruptions in these consecutive segments, resulting in a boundary point for sign change. The number and length of consecutive segments with the same sign in the difference sequence are then counted. A consecutive segment with the same sign refers to a continuous sequence where the sign of the difference value remains unchanged. The statistical process traverses the entire difference sequence, recording the starting position and length of each consecutive segment with the same sign. The overall monotonicity score of the sub-segment data is calculated based on the distribution characteristics of consecutive segments with the same sign. These characteristics include the total number of consecutive segments with the same sign, their average length, and the length of the longest consecutive segment. The monotonicity score is derived by weighting these characteristic values; a higher score indicates stronger monotonicity in the sub-segment. The absolute value of the difference in monotonicity between the forward and backward sub-segments serves as a trend consistency indicator; a smaller absolute value indicates stronger consistency in the changing trends of the two sub-segments. In some embodiments, the calculation of the monotonicity score can incorporate an auxiliary indicator based on the consistency of derivative signs to enhance adaptability to complex fluctuation patterns.

[0031] In practical implementation, the fluctuation coordination index is calculated by analyzing the periodic characteristics of the data fluctuations in both the forward and backward sub-segments. The periodic characteristics refer to the main time scale of the data fluctuations, extracted using a period estimation algorithm. The matching degree of the fluctuation periods of the two sub-segments is calculated as the fluctuation coordination index, measuring the similarity between the fluctuation periods of the forward and backward sub-segments. The periodic characteristics analysis employs Fast Fourier Transform (FFT) or Autocorrelation Analysis (ASA). FFT transforms the sub-segment data from the time domain to the frequency domain, identifying the dominant frequency component; the period is derived from the reciprocal of the dominant frequency. ASA calculates the correlation between the sub-segment data and its delayed replica, determining the period by finding the peak position of the ASA function. The matching degree is calculated based on the ratio or difference between the forward and backward sub-segment period values; a ratio close to 1 or a difference close to 0 indicates a high matching degree. A high fluctuation coordination index value indicates that the fluctuation patterns of the forward and backward sub-segments are coordinated. In some embodiments, period analysis can be combined with multi-resolution analysis to evaluate fluctuation characteristics at different time scales, improving the robustness of period estimation.

[0032] Optionally, the calculation of monotonicity can integrate nonlinear trend detection technology. Before differencing, multinomial fitting is performed on the sub-segment data to separate trend and fluctuation components, thus more accurately assessing monotonicity. It is understood that the statistics of consecutive segments with the same sign need to consider the amplitude threshold of the difference value to ignore sign changes caused by minor fluctuations. In specific implementation, the calculation process of the trend consistency index and fluctuation coordination index is embedded in the real-time data processing loop. Each feature data segment is processed independently, and the index values ​​are stored as vectors for subsequent fusion. The weight parameters of the monotonicity score are determined through offline calibration, which uses historical data segments labeled with ideal monotonicity for optimization. The calculation of fluctuation cycle matching degree supports multi-cycle matching. When multiple significant cycles exist in a sub-segment, the matching degree is based on the weighted average of the matching results of each cycle.

[0033] In practical implementation, the segmentation of forward and backward sub-segments handles boundary cases. When the maximum value is located at the endpoint of a data segment, a sub-segment may be empty. In this case, adjacent data points are inserted or the entire data segment is used as a single sub-segment. Differential processing employs first-order forward differencing, with the length of the differencing sequence being one less than the number of data points in the atomic segment. The statistical analysis of consecutive segments with the same sign uses a sliding window to verify sign consistency, avoiding transient sign flips caused by noise. The monotonicity score calculation formula integrates the reciprocal of the number of consecutive segments with the same sign, the proportionality of the average length, and the proportionality of the maximum length, ensuring that long but few consecutive segments receive high scores. The output of the trend consistency index is normalized, mapping the absolute value of the difference to the interval between 0 and 1 for easier subsequent comparison. In the fluctuation period analysis, the parameters of the Fast Fourier Transform, such as the window function and length, are adaptively selected based on the number of sub-segment data points to ensure appropriate frequency resolution. The calculation of the period matching degree introduces a tolerance mechanism, allowing period values ​​to fluctuate within a small range while still being considered a match. The tolerance size is set based on the trolley motion characteristics.

[0034] In practical implementation, the calculation of the fluctuation coordination index can be extended to multi-dimensional coordination analysis. When the pose data contains multiple parameters such as x, y, and z directional positions, the fluctuation period of each parameter is calculated separately, and the overall coordination index is derived by combining the matching degree of each parameter. The extraction of periodic characteristics is performed independently for each sub-segment. When the sub-segment length is insufficient, zero padding or interpolation methods are used to meet the minimum data requirements of the periodic analysis algorithm. The matching degree result is converted into a binary coordination flag through threshold judgment, but continuous values ​​are usually retained for fine evaluation. It can be understood that the trend consistency index and the fluctuation coordination index jointly characterize the symmetry and regularity of the data segment changes, and the output of the index calculation module is directly passed to the periodic regularity coefficient generation module.

[0035] In practical implementation, the optimization of monotonicity calculation involves smoothing the difference sequence preprocessing. Smoothing employs a moving average filter, with the filter window size correlated to the data sampling frequency to suppress high-frequency noise. Statistical algorithms for consecutive segments with the same sign optimize computational efficiency by dynamically updating the consecutive segment information in a single pass through the difference sequence, reducing memory usage. The weighting coefficients for the monotonicity score are adjusted using machine learning methods, with training data derived from manually labeled high- and low-monotonic sub-segments. The calculation of the trend consistency index considers the normalization of sub-segment length differences to avoid bias in comparing sub-segments of varying lengths. In fluctuation cycle analysis, parabolic interpolation is used to improve the accuracy of peak detection in the autocorrelation function, avoiding cycle quantization errors caused by sampling intervals. Cycle matching degree calculation supports weighted matching, with weights set based on the amplitude of the cycle components; cycle components with larger amplitudes contribute greater weights.

[0036] In practical implementation, the real-time performance of the index calculation is improved by parallel processing of the forward and backward segments. Parallel computing leverages the thread capabilities of multi-core processors to simultaneously perform monotonicity evaluation and periodic analysis for both segments. Sign determination of the difference sequence is accelerated using vectorized operations, fully utilizing the processor's SIMD instruction set. The choice of periodic analysis algorithm is based on a trade-off between accuracy and speed; in resource-constrained embedded systems, computationally lighter autocorrelation analysis is prioritized. The output of the volatility consistency index is supplemented with a confidence score, calculated based on the spectral purity of the period estimate or the significance of the autocorrelation peak. The entire index calculation process is encapsulated as an independent software module. The module takes a set of feature data segments as input and outputs arrays of trend consistency index values ​​and volatility consistency index values ​​for each data segment.

[0037] In practical implementation, the calculation modules for trend consistency and fluctuation coordination indicators incorporate error handling. When sub-segment data is invalid or calculation fails, the indicator value is set to a default value or a recalculation is triggered. The module's performance monitoring records calculation time and indicator value distribution for system diagnosis and optimization. The parameters of the monotonicity scoring formula provide an external configuration interface, allowing adjustments to the scoring strategy based on different tunnel engineering conditions. The tolerance parameter for periodic matching is dynamically adjusted, adaptively narrowing or widening the tolerance range based on the stability of historical matching results. The indicator calculation process log records detailed intermediate results, such as difference sequences and periodic values, for offline analysis and algorithm verification. The module exchanges data with upstream and downstream data processing steps via message queues or shared memory to ensure continuous data flow.

[0038] Consider a specific application scenario: a tunnel lining trolley performs concrete lining work within a straight tunnel. The trolley moves slowly along the tunnel axis at a constant speed. Simultaneously, due to minor unevenness in the tunnel floor or slight pressure fluctuations in the hydraulic system, the vertical (Z-axis) position measurement of the trolley will exhibit regular fluctuations as reflected in the sensors. A multi-sensor system collects a sequence of pose data of the trolley's Z-axis position over a period of time. After multi-scale segmentation processing as described in Example 1, a set of characteristic data segments is identified, including a representative segment for analysis. This data segment encompasses the fluctuations in the vertical position of the trolley throughout a complete work cycle. The segment begins at a trough (lower position), rises to a peak (highest point), and then falls back to another trough.

[0039] In practical implementation, analyzing the change pattern of this characteristic data segment first requires locating the maximum value point within the data segment, i.e., the highest vertical position reached by the trolley in this work cycle. By traversing and comparing the values ​​of all data points within the data segment, the specific location index of the maximum value point can be determined. Using this maximum value point as the dividing point, the original data segment is clearly divided into two sub-segments: the forward sub-segment, from the initial valley point to the maximum value point, includes the stage of the trolley's vertical position rising; the backward sub-segment, from the maximum value point to the final valley point, includes the stage of the trolley's vertical position falling. This segmentation method helps to independently examine the motion characteristics of the rising and falling stages.

[0040] Calculating the trend consistency index requires assessing the monotonicity of the data trends in the forward and backward sub-segments separately. For the forward sub-segment, the ideal trend should be monotonically increasing. To calculate its monotonicity, the consecutive data points within the sub-segment are differencinged; that is, the position value of the next point is subtracted from the position value of the previous point, resulting in a difference value sequence. In an ideal, steady upward trend, all these difference values ​​should be positive. The number and length of consecutive segments with the same sign in this difference sequence are counted. If all the difference values ​​in the entire forward sub-segment are positive, there is only one consecutive segment with the same sign, and its length is equal to the number of data points in the sub-segment minus one, indicating a high degree of monotonicity. Similarly, for the backward sub-segment, the ideal trend is monotonically decreasing, and all difference values ​​should be negative. The monotonicity of the backward sub-segment is calculated by counting consecutive segments with the same sign in its negative difference sequence. After obtaining the monotonicity scores for the forward and backward sub-segments, the trend consistency index is calculated as the absolute value of the difference between these two scores. In this ideal example, if both the ascent and descent phases are ideal, with high and similar monotonicity scores, the absolute difference is small, indicating high trend consistency—that is, good trend symmetry between the ascent and descent. However, if uneven ground causes fluctuations during the ascent (brief descent) and fluctuations during the descent (brief ascent), the monotonicity scores for both segments will decrease, and the scores may differ significantly, leading to an increase in the trend consistency index and reflecting poor consistency in the motion pattern.

[0041] The calculation of the volatility coordination index focuses on the periodic characteristics of data fluctuations in the forward and backward segments. Even if the overall trend is upward or downward, data points may fluctuate around the main trend line due to mechanical vibrations or measurement noise. To analyze the periodic characteristics of the forward segment's data fluctuations, the data in that segment can be detrended (e.g., by subtracting a linearly fitted line), and then autocorrelation or spectral analysis can be performed on the residual sequence to find its main fluctuation period. Similarly, the detrended data of the backward segment is analyzed to extract its main fluctuation period. The matching degree of the fluctuation periods of these two segments is calculated; for example, the ratio of the two period values ​​is calculated. If the ratio is close to 1, the matching degree is high, indicating that the fluctuation rhythm of the upward phase is coordinated with the fluctuation rhythm of the downward phase. The volatility coordination index value is determined by this matching degree. In this example, if the trolley hydraulic system is stable and the support is uniform, then the small fluctuation periods in the upward and downward phases may be similar, resulting in a good coordination index. If a local obstacle is encountered during the upward phase, causing vibrations at a specific frequency, but not during the downward phase, the fluctuation periods of the two segments may differ significantly, resulting in a worse coordination index.

[0042] In practice, the calculation details of monotonicity require careful handling. For example, when statistically analyzing consecutive segments with the same sign in a difference sequence, a small tolerance threshold needs to be set to avoid frequent sign flipping due to numerical calculation errors or extremely small fluctuations. The distribution characteristics of consecutive segments with the same sign, such as the total number of segments (fewer segments indicate better trend consistency), the average length (longer length indicates a more stable trend), and the maximum segment length, are all used to calculate a comprehensive monotonicity score. The scoring function may assign higher weights to the average length and the maximum length.

[0043] For analyzing the periodic characteristics of fluctuations, autocorrelation analysis is a commonly used method. It identifies periodicity by calculating the correlation coefficient between detrended data and its delayed replicas. The delay time corresponding to the peak point in the correlation coefficient sequence represents the potential fluctuation period. Spectral analysis (such as Fast Fourier Transform) can transform the data from the time domain to the frequency domain; the period corresponding to the frequency component with the highest energy is the primary fluctuation period. Calculating the degree of matching may require considering multiple periodic components, such as calculating the weighted average degree of matching between the primary and secondary periods.

[0044] This specific example scenario demonstrates how, starting from a data segment representing the vertical positional fluctuations of a tunnel lining trolley, the characteristics of its rising and falling phases in terms of trend consistency and fluctuation coordination can be quantified through segmentation, differencing, and statistical analysis. The calculated trend consistency and fluctuation coordination indices provide crucial quantitative input for subsequent evaluation of the trolley's motion regularity reflected in this data segment. The entire calculation process can be automatically executed by an algorithm program on an embedded system, processing the continuously arriving sensor data in real time.

[0045] See Figure 4 This graph focuses on analyzing the variation patterns of each data segment within a set of characteristic data segments, and the technical steps involved in calculating trend consistency and fluctuation coordination indices. The horizontal axis represents the characteristic data segment number, and the vertical axis represents the index value. Red bars represent trend consistency indices, and orange bars represent fluctuation coordination indices. The graph is generated based on a technical process: dividing the data segment into forward and backward sub-segments by the maximum value point; calculating the difference in monotonicity between sub-segments to obtain the trend consistency index; and analyzing the fluctuation cycle matching degree of the sub-segments to obtain the fluctuation coordination index. Its data provides crucial quantitative support for subsequent generation of periodicity coefficients, evaluation of pose data quality, and dynamic adjustment of control strategies, and is of great significance for achieving precise control of the pose of tunnel lining trolleys.

[0046] Example 3: In specific implementation, the extraction of the extreme value evolution pattern description starts with all extreme points in the feature data segment set arranged in chronological order, constructing an extreme point sequence. This sequence contains all local peaks and valleys extracted from the feature data segment set, arranged in ascending order of timestamps to form an ordered sequence. Constructing the extreme point sequence requires processing boundary extreme points between multiple feature data segments to ensure chronological continuity and prevent duplicate counting. The numerical change gradient sequence of adjacent extreme points in the extreme point sequence is calculated. The numerical change gradient is obtained by dividing the numerical difference between adjacent extreme points by the corresponding time interval. The numerical difference reflects the magnitude change, and the time interval reflects the rate of change. The numerical change gradient sequence constitutes a vector reflecting the dynamic changes between extreme points. The stability characteristics of the numerical change gradient sequence are analyzed. These stability characteristics are quantified by calculating the statistical dispersion of the numerical change gradient sequence, for example, using the standard deviation or coefficient of variation. The coefficient of variation is the ratio of the standard deviation to the mean, which can eliminate the influence of dimensions and facilitate comparison. The distribution pattern of time intervals between statistical extreme points is analyzed. A time interval refers to the time difference sequence between adjacent points in the extreme point sequence. The coefficient of variation (COP) of the time intervals is calculated, and it is also expressed as the ratio of the standard deviation to the mean, used to measure the degree of fluctuation in the time intervals. The stability characteristics of the numerical change gradient sequence and the COP of the time intervals are combined to form an extreme value evolution pattern description. This combination can be achieved by constructing a two-dimensional feature vector or by weighted fusion into a scalar descriptor. The extreme value evolution pattern description characterizes the rhythm and regularity of extreme point evolution. In some embodiments, trend analysis can be introduced into the analysis of the numerical change gradient sequence to examine the trend of the numerical change gradient itself over time, in order to describe the evolution pattern in more detail.

[0047] In practical implementation, the generation of the periodicity coefficient first involves calculating the weighted average of the trend consistency index and fluctuation coordination index of all data segments in the feature data segment set. The weights in the weighted average can be allocated based on the length or variation range of each data segment; longer or larger data segments are usually assigned higher weights to emphasize their importance. Calculating the weighted average requires normalizing the trend consistency index and fluctuation coordination index values ​​of all data segments to ensure they are within a comparable range. Next, the morphological similarity between data segments in the feature data segment set is analyzed, and the average similarity score is calculated. Morphological similarity assessment is achieved by comparing the shape characteristics of the data segments; for example, a dynamic time warping algorithm can be used to calculate the warped path distance between each pair of data segments. This distance is then converted into a similarity score, and finally, the average similarity score of all data segment pairs is calculated as the average similarity score. The stability features and coefficient of variation in the extreme value evolution pattern description are normalized to obtain the extreme value change regularity. Normalization maps the stability feature values ​​and coefficient of variation values ​​to the interval between 0 and 1. The extreme value change regularity can be expressed as a function of the complementary values ​​of the normalized stability feature values ​​and the normalized coefficient of variation values, such as the average or product. The weighted average, average similarity score, and extreme value change regularity are then fused to generate a periodicity coefficient. This fusion calculation typically uses weighted summation or geometric mean. A higher periodicity coefficient indicates stronger periodicity and better regularity in the pose data sequence. In some embodiments, the average similarity score can be calculated using a shape feature extraction method based on principal component analysis to reduce computational complexity.

[0048] Optionally, the construction of the extreme point sequence can include labeling of the extreme point type (peak or valley), and distinguish between peak-to-valley and valley-to-peak change patterns when calculating the numerical change gradient, thereby generating a more refined description of the evolution pattern. It is understood that the stability analysis of the numerical change gradient sequence may require excluding obviously anomalous gradient values. These outliers may originate from transient sensor interference and need to be filtered out using statistical outlier detection methods. In specific implementation, the generation process of the periodic regularity coefficients can be iterative, dynamically updating the coefficient values ​​using newly arrived feature data segments to adapt to the slow changes in the tunnel lining trolley's movement pattern.

[0049] In practical implementation, the extraction process of the extreme value evolution pattern description involves in-depth statistical analysis of the numerical change gradient sequence and the time interval sequence. The stability characteristics of the numerical change gradient sequence of adjacent extreme points in the extreme point sequence can be quantified using the following formula: .in, Indicators representing the characteristics of gradient stability indices for numerical changes The standard deviation of the gradient sequence of numerical changes This represents the arithmetic mean of the gradient sequence of numerical changes. It is a very small constant used to avoid division by zero errors. Stability index characteristics. The closer the value is to 1, the more stable the numerical gradient sequence. Characteristics of the coefficient of variation over time intervals. The calculation formula is .in, It is the standard deviation of the time interval sequence. It is the arithmetic mean of the time interval sequence. The extreme value evolution pattern description can ultimately be represented as a stability feature containing the changing gradient sequence. Characteristics of the coefficient of variation of time intervals tuple In some embodiments, it is also possible to and The reciprocal of the product or other functions are combined to form a comprehensive regularity score.

[0050] In practical implementation, the fusion calculation of the periodicity coefficient requires determining the relative weights of three components: the weighted average, the average similarity score, and the regularity of extreme value changes. These weights can be learned from historical data or set as fixed values ​​based on engineering experience. For example, a feasible fusion formula is as follows: .in, Represents the periodic regularity coefficient of the final generated periodicity. This represents the weighted average of trend consistency indicators and volatility coordination indicators. The average similarity score represents the set of feature data segments. This represents the regularity of extreme value changes derived from the extreme value evolution model. It is a weighting coefficient and satisfies The weighting coefficients can be set with a bias towards the average similarity score. This is because the similarity in the shape of data segments directly reflects periodicity. This is because the similarity in the shape of data segments can intuitively reflect the repetitive characteristics during the trolley's movement, and this repetitive characteristic is the core basis for judging whether a pose data sequence possesses periodicity. Compared to local indicators such as trend consistency, fluctuation coordination, and the regularity of extreme value changes, shape similarity is more decisive in representing the overall periodicity. Therefore, it is given a higher weight in the weight allocation to highlight its dominant role in the calculation of the periodicity coefficient. It can be understood that, in the specific implementation, all input components... , and All of them need to be normalized to ensure that their dimensions are consistent and within the same numerical scale (such as the 0-1 range) before weighted fusion.

[0051] In practical implementation, the construction of the extreme point sequence must ensure strict accuracy in time order; any timestamp error may lead to distortion of the evolutionary pattern description. The calculation of numerical gradients employs double-precision floating-point arithmetic to guarantee accuracy, and the calculation of time intervals also requires high-precision timer data. To avoid division-by-zero errors in the stability feature calculation and to prevent this constant from substantially interfering with the calculation results, the minimum constant in the stability feature calculation formula is... Typically set to a value much smaller than typical. The value of the constant. In the calculation of the numerical gradient stability index, if the arithmetic mean of the numerical gradient sequence approaches zero, there will be invalid calculations with a denominator of zero. Adding this minimal constant can ensure the validity of the fractional operation. Setting it to a value much smaller than the typical mean of the numerical gradient sequence can minimize the influence of the constant itself on the stability index calculation results, ensuring that the index can truly reflect the discrete characteristics of the numerical gradient sequence, thereby accurately characterizing the stability of the extreme point evolution and providing reliable data support for the subsequent calculation of periodicity coefficients and quality assessment coefficients.

[0052] Dynamic time warping algorithms in morphological similarity calculations require selecting an appropriate step size and may involve downsampling to control computational costs. (Extremum variation regularity) The calculation can be designed as follows: .in, It is the coefficient of variation of the normalized time interval. Periodicity coefficient. The calculation results are cached for comparison with coefficients calculated in subsequent time windows to monitor the system's periodic trends. The entire calculation process is encapsulated as a callable module; the input is a set of feature data segments and a list of extreme points, and the output is the periodicity coefficient. This coefficient will be used in the calculation of the quality assessment coefficient.

[0053] See Figure 5 This figure focuses on the technical aspect of generating the periodicity coefficient of the pose data sequence. The horizontal axis represents the feature data segment number, and the vertical axis represents the periodicity coefficient. The periodicity coefficient is a quantitative indicator generated by integrating multi-dimensional information such as trend consistency, fluctuation coordination, morphological similarity, and extreme value evolution patterns of the feature data segments. It is used to characterize the periodicity and stability of the trolley's motion. This coefficient is one of the core bases for subsequent calculation of quality evaluation coefficients and dynamic adjustment of pose processing strategies. Its upward trend indicates that the periodicity of the trolley's motion is gradually becoming more prominent, providing key quantitative support for achieving precise and stable control of the tunnel lining trolley's pose. It helps technicians judge the predictability of the trolley's motion pattern, thereby optimizing the adaptability and accuracy of the control algorithm.

[0054] Example 4: In specific implementation, the calculation of the concentration and dispersion indices of the feature components begins with singular value decomposition (SVD) of the spatiotemporal feature matrix. The spatiotemporal feature matrix is ​​composed of pose data sequences organized in chronological order, with rows corresponding to different time points and columns corresponding to different dimensional parameters of the pose (such as X, Y, Z coordinates and pitch and yaw angles). SVD decomposes the spatiotemporal feature matrix into the product of three matrices: a left singular vector matrix, a diagonal singular value matrix, and the transpose of a right singular vector matrix. The singular values ​​in the diagonal singular value matrix are arranged in descending order to form a singular value spectrum, which reflects the energy distribution of the information contained in the spatiotemporal feature matrix. An adaptive thresholding algorithm is used to divide the singular value spectrum into principal component regions and secondary component regions. The algorithm dynamically determines the partitioning point based on the cumulative contribution rate of singular values. For example, the singular value index at which the cumulative contribution rate first reaches or exceeds a set threshold (e.g., 90% of the total variance) is used as the partition boundary. All singular values ​​before this index constitute the principal component region, and those after constitute the secondary component region. The cumulative contribution rate of singular values ​​in the principal component region is calculated as a concentration index. The cumulative contribution rate is the ratio of the sum of squares of all singular values ​​in the principal component region to the sum of squares of all singular values ​​in the entire region. This index measures the degree to which most of the information is concentrated in the principal component. The degree of difference in the distribution of singular values ​​between the principal and secondary component regions is analyzed as a dispersion index. The degree of difference can be quantified by calculating the ratio of the sum of squares of singular values ​​in the two regions, the logarithmic ratio, or a measure based on information entropy. The dispersion index reflects the dispersion of information distribution between the principal and secondary components.

[0055] In practical implementation, the calculation of the quality assessment coefficient first requires calculating the correlation matrix between the row vectors of the spatiotemporal feature matrix. The row vectors represent the pose states at different time points. The correlation matrix is ​​obtained by calculating the Pearson correlation coefficient between each pair of row vectors, forming a symmetric square matrix. The internal structural features of the correlation matrix are analyzed. These features can be obtained by examining the eigenvalue distribution, the rank of the matrix, or by applying clustering algorithms to identify the grouping patterns of the row vectors. For example, an ideal high-quality sequence might exhibit a block diagonal structure or a high eigenvalue decay rate in its correlation matrix. The analyzed internal structural features are then coupled with the periodicity coefficient to obtain the sequence's regularity score. This coupling calculation can be a simple linear weighted combination or a nonlinear mapping, such as multiplying the periodicity coefficient by the principal eigenvalues ​​or eigenvalue entropy of the correlation matrix. Finally, the concentration and dispersion indices of the feature components are combined and transformed to obtain the sequence's feature integrity score. This transformation typically involves mathematical operations on the two indices, such as weighted summation, product, or scaling one index on the other. The feature integrity score aims to assess whether the features extracted from matrix decomposition sufficiently and compactly represent the original sequence. The system combines regularity scores and feature integrity scores, and generates a final quality assessment coefficient through weighted fusion. The weights can be preset or adaptively adjusted based on the relative importance of regularity and feature integrity in specific application scenarios. The quality assessment coefficient, as a scalar output, is used for dynamic adjustment of subsequent pose data processing strategies.

[0056] Optionally, the singular value decomposition (SVD) calculation can employ a truncated SVD algorithm to improve the processing efficiency of large-scale matrices, especially when the number of rows and columns of the spatiotemporal feature matrix is ​​large. It can be understood that the cumulative contribution rate threshold in the adaptive thresholding algorithm can be fine-tuned according to the specific motion pattern of the tunnel lining trolley. For trolleys with more complex motion patterns, the threshold can be appropriately lowered to capture more nuanced features. In practical implementation, the correlation matrix analysis can further calculate its condition number or determinant value, which can then be used as an additional indicator of internal structural stability in the calculation of regularity scoring.

[0057] In practical implementation, the calculation process of concentration and dispersion indices requires stability handling of numerical calculations. For example, when calculating the cumulative contribution rate, the case where the denominator (sum of squares of all singular values) is close to zero needs to be avoided. This can be ensured by checking the norm of the matrix. The calculation formula for the dispersion index should be able to distinguish between highly concentrated information distribution (high concentration, low dispersion) and relatively uniform information distribution (relatively low concentration, high dispersion). The generation process of the quality assessment coefficient can be designed as a configurable computation graph, where the parameters (such as weights and thresholds) of each computation step (such as matrix factorization, correlation calculation, and index fusion) can be set through a configuration file to adapt to different tunnel engineering environments. The fusion function of regularity score and feature integrity score needs to be normalized so that the final quality assessment coefficient falls within a fixed range (such as between 0 and 1), facilitating comparison with preset thresholds for decision-making. Refer to Table 1, which shows the typical relationship between concentration index, dispersion index, and feature integrity score under different feature distribution patterns. It should be noted that the feature integrity score is the result of the combined effect of the concentration index and the dispersion index, and the specific calculation formula can be defined according to actual needs.

[0058] Table 1. Relationship between Feature Distribution Patterns and Indicators

[0059] In some embodiments, the construction of the spatiotemporal feature matrix can incorporate time delay embedding to capture dynamic characteristics. That is, each row contains not only the pose at the current moment but also the pose data from the previous few time points, thus expanding the matrix to a form such as a Hankel matrix. The correlation matrix can also be calculated using other types of correlation coefficients, such as the Spearman rank correlation coefficient, to reduce sensitivity to outliers. Besides weighted averaging, the fusion of regularity scores and feature integrity scores can also employ rule-based methods. For example, when the regularity score is below a certain threshold, regardless of how high the feature integrity score is, the final quality assessment coefficient is forced to a lower value.

[0060] In practical implementation, the entire computation process requires the support of efficient linear algebra libraries, such as using the interfaces of BLAS and LAPACK libraries for singular value decomposition and correlation matrix calculations, to ensure computational speed when processing high-dimensional pose data. For scenarios with high real-time requirements, incrementally updated singular value decomposition algorithms can be used to avoid decomposing the complete spatiotemporal feature matrix each time. The results of concentration and dispersion indices change dynamically with the arrival of new data, therefore, the quality assessment coefficients need to be recalculated periodically. The output of the quality assessment coefficient calculation module serves as a key input to the dynamic adjustment module of the pose data processing strategy, thus forming a closed-loop adaptive control system. The implementation of this module is typically contained within the software layer of the embedded system, requiring careful management of memory usage and computational load to meet the performance constraints of the real-time control system.

[0061] Example 5: In specific implementation, the dynamic adjustment method of pose data processing strategy determines the intensity parameter of pose data filtering based on the quality assessment coefficient, establishing a mapping relationship between the filtering intensity and the quality assessment coefficient. This mapping relationship is usually an inverse proportional function. That is, when the quality assessment coefficient is high, it indicates that the pose data sequence itself has good regularity and little noise interference. Therefore, a lower filtering intensity is selected to preserve the true dynamic characteristics of the data. When the quality assessment coefficient is low, it indicates that the data sequence is subject to greater interference or irregular fluctuations. In this case, a higher filtering intensity is used to suppress noise and improve the smoothness and stability of the data. The filtering intensity parameter can specifically refer to the filter's cutoff frequency, window length, or smoothing factor, etc. For example, when using a low-pass filter, the cutoff frequency is inversely proportional to the quality assessment coefficient. The mapping relationship can be established through a preset lookup table or a simple mathematical function. The lookup table contains a series of quality assessment coefficient intervals and their corresponding recommended filtering intensity parameter values, while the mathematical function can be a linear decreasing function or a nonlinear saturating function. Based on the filtered pose data, the real-time pose state of the tunnel lining trolley is calculated. This calculation involves parsing the filtered data sequence, which may include obtaining velocity and displacement from acceleration data through integration, or calculating the trolley's precise position and attitude angles in three-dimensional space using a fusion algorithm combined with multi-sensor data (such as GPS, IMU, and laser rangefinder). Based on the deviation between the real-time pose state and the desired pose, a pose adjustment command is generated. The desired pose is typically given by the tunnel design model or a preset trajectory. The deviation calculation involves solving for the position error vector and attitude error angle. The pose adjustment command contains specific control quantities, such as the hydraulic cylinder stroke, motor speed, or wheel steering angle to be adjusted. The pose adjustment command is executed through a closed-loop control system to achieve real-time control of the trolley's pose. The closed-loop control system receives the pose adjustment command, drives the actuators (such as hydraulic servo valves, electric push rods, and steering motors), and simultaneously monitors pose changes in real time through sensor feedback, forming a feedback loop to continuously correct deviations until the desired pose is achieved.

[0062] In some embodiments, the mapping relationship between filter strength and quality assessment coefficients can be designed as a function with hysteresis characteristics to prevent system oscillations caused by frequent switching of filter strength near the critical value of the quality assessment coefficients. It is understood that calculating the real-time pose state may require considering the kinematic model of the tunnel lining trolley, and inputting the filtered sensor data into the model for state estimation to improve the accuracy of pose calculation.

[0063] Optionally, the choice of data filtering algorithm can be adapted to different quality assessment coefficient levels. For example, when the quality assessment coefficient is extremely low, in addition to increasing the filtering strength, a more robust filtering algorithm, such as median filtering or H-infinity filtering, can be switched to cope with impulse noise or model uncertainty. Optionally, predictive control algorithms can be introduced into the generation of pose adjustment commands, which not only consider the pose deviation at the current moment but also predict the deviation trend in the next few steps, thereby generating smoother and more forward-looking control commands. Optionally, the controller parameters of the closed-loop control system (such as the proportional, integral, and derivative coefficients of the PID controller) can also be correlated with the quality assessment coefficient. When the quality assessment coefficient is high, more aggressive control parameters can be used to speed up the response. When the quality assessment coefficient is low, more conservative control parameters are used to ensure system stability.

[0064] In practical implementation, establishing the mapping relationship between filter intensity and quality evaluation coefficients requires a calibration process. This calibration is performed in the early stages of system deployment by injecting simulated noise of varying intensities and observing the pose estimation error under different filter intensities, thereby determining the optimal mapping relationship for the specific tunnel lining trolley. The real-time pose state calculation module needs to efficiently process the data stream, typically employing a real-time operating system on an embedded system to ensure timely completion of the calculation tasks. The logic for generating pose adjustment commands needs to include safety constraint checks, such as checking whether the command will cause the trolley to collide with the tunnel wall or exceed the working range of the actuator. The execution cycle of the closed-loop control system needs to be strictly synchronized with the data acquisition cycle to ensure the timeliness and accuracy of control. The entire dynamic adjustment process constitutes a nested control loop; the inner loop is the traditional pose closed-loop control, and the outer loop is a parameter adaptive adjustment loop based on the quality evaluation coefficients.

[0065] In some embodiments, the quality assessment coefficient itself can also be introduced as a feedback signal into the generation logic of the pose adjustment command. For example, when the quality assessment coefficient remains consistently low, it may indicate a sensor malfunction or a drastic environmental change, which can trigger a degraded operating mode or an alarm to prompt operator intervention. It is understood that the adjustment of the filter intensity needs to be smooth to avoid abrupt changes in pose control caused by step changes in parameters.

[0066] In practical implementation, data filtering may employ a Kalman filter or its variants. The filtering strength is achieved by adjusting the process noise covariance matrix and observation noise covariance matrix of the Kalman filter. A lower quality assessment coefficient corresponds to a larger setpoint for the process noise covariance matrix, indicating a lower confidence in the model's predictions and a greater reliance on observation data, but also stronger smoothing of the observation data. Real-time pose state calculation may involve complex coordinate transformations, converting data from the local sensor coordinate system to the global tunnel design coordinate system. The calculated pose deviation is typically a multidimensional error vector, which needs to be mapped to specific control variables. For example, a control allocation algorithm can be used to decompose the overall pose error into action commands for each independent actuator (such as multiple support cylinders). The actuator response characteristics of the closed-loop control system need to be accurately modeled, and their dynamic delays and nonlinearities need to be considered when generating commands. Sometimes, feedforward compensation is required to improve tracking performance. The code implementation of the entire dynamic adjustment method needs to undergo rigorous testing, including unit testing, integration testing, and hardware-in-the-loop testing, to ensure reliability in a real tunnel environment.

[0067] In practical implementation, the real-time performance of the dynamic adjustment strategy is crucial. Therefore, all calculation steps require optimized algorithm complexity. For example, binary search is used for mapping relationship lookup, and iterative filtering algorithms are employed to avoid redundant calculations. The system needs to continuously monitor the changing trend of the quality assessment coefficients. If a rapid decrease in the quality assessment coefficients is detected within a short period, it may indicate a sudden disturbance. In this case, the filtering intensity setting based on the current coefficients can be temporarily overridden, and the highest intensity filtering mode can be used directly to address the issue. The issuance of pose adjustment commands requires a reliable communication protocol to ensure that the commands are delivered to the actuators accurately and promptly. Simultaneously, the system needs to record detailed operation logs, including historical data on quality assessment coefficients, filtering intensity parameters, pose deviations, and control commands, for subsequent performance analysis and fault diagnosis. Through this dynamic adjustment method, the pose control system of the tunnel lining trolley can adapt to complex tunnel construction environments and maintain high-precision pose control capabilities under various data quality conditions.

[0068] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time regulation of a tunnel lining jumbo pose based on multi-sensor data, characterized in that, The method comprises: Collecting multi-dimensional pose measurement data of a tunnel lining trolley during movement, and constructing a time-space correlated pose data sequence; Performing multi-scale segmentation processing on the pose data sequence, and identifying a characteristic data segment set reflecting the movement law of the trolley; Analyzing the change mode of each data segment in the characteristic data segment set, and calculating a trend consistency index and a fluctuation coordination index of the data inside each data segment; Detecting the extreme point distribution law in the characteristic data segment set, extracting the change gradient and time interval features between the extreme points, and forming an extreme evolution mode description; Comprehensively considering the overall distribution characteristics of the characteristic data segment set, the extreme evolution mode description, and the change mode of each data segment, generating a periodic regularity coefficient of the pose data sequence based on the trend consistency index and the fluctuation coordination index; Constructing a time-space feature matrix of the pose data sequence, extracting main feature components of the sequence through matrix decomposition technology, and calculating a concentration index and a dispersion index of the feature components; Fusing the internal correlation features of the time-space feature matrix, the periodic regularity coefficient of the pose data sequence, and the concentration index and dispersion index of the feature components to obtain a quality evaluation coefficient of the pose data sequence; Based on the quality evaluation coefficient, dynamically adjusting the pose data processing strategy to realize the pose regulation and control of the tunnel lining trolley.

2. The real-time pose regulation method of a tunnel lining jumbo based on multi-sensor data according to claim 1, characterized in that, The multi-scale segmentation processing on the pose data sequence to identify the characteristic data segment set reflecting the movement law of the trolley comprises: Using a sliding window detection algorithm to scan the pose data sequence, identifying all extreme points in the sequence, including local peak points and valley points; Dividing the pose data sequence into multiple continuous data segments with adjacent valley points as boundaries; Calculating the data change amplitude of each data segment, and grouping the change amplitudes of all data segments through a density clustering algorithm; Selecting the data segment group with the largest change amplitude as the characteristic data segment set.

3. The real-time pose regulation method for a tunnel lining jumbo based on multi-sensor data according to claim 1, characterized in that, The calculation method of the trend consistency index and the fluctuation coordination index comprises: For each data segment in the characteristic data segment set, taking the maximum point of the data segment as a dividing point, dividing the data segment into a forward sub-segment and a backward sub-segment; Calculating the monotonicity degree of the data change trend of the forward sub-segment and the monotonicity degree of the data change trend of the backward sub-segment, and taking the absolute value of the difference between the two monotonicity degrees as the trend consistency index; Analyzing the periodic characteristics of the data fluctuation of the forward sub-segment and the data fluctuation of the backward sub-segment, calculating the matching degree of the fluctuation periods of the two sub-segments, and taking the matching degree of the fluctuation periods of the two sub-segments as the fluctuation coordination index.

4. The real-time pose regulation method of a tunnel lining jumbo based on multi-sensor data according to claim 3, characterized in that, The calculation method of the monotonicity degree comprises: Performing difference processing on the sub-segment data, and counting the number and length of the same sign continuous segments in the difference sequence; According to the distribution characteristics of the same sign continuous segments, calculating the overall monotonicity score of the sub-segment data to obtain the monotonicity degree.

5. The real-time pose regulation method for a tunnel lining jumbo based on multi-sensor data according to claim 2, characterized in that, The extraction method of the extreme evolution mode description comprises: Arranging all extreme points in the characteristic data segment set in time sequence to construct an extreme point sequence; Calculating the numerical change gradient sequence of adjacent extreme points in the extreme point sequence, and analyzing the stability features of the change gradient sequence; Statistically analyzing the time interval distribution law between the extreme points, and calculating the coefficient of variation of the time interval; The stability characteristics of the changing gradient sequence and the coefficient of variation of the time interval are combined to form an extreme value evolution pattern description.

6. The multi-sensor data based real-time pose regulation method for a tunnel lining jumbo according to claim 1, characterized in that, The method for generating the periodicity coefficient includes: Calculate the weighted average of the trend consistency index and volatility coordination index for all data segments in the feature data segment set; Analyze the morphological similarity between data segments in the feature data segment set, and calculate the average similarity score of the feature data segment set; The stability characteristics and coefficient of variation in the description of extreme value evolution patterns are normalized to obtain the regularity of extreme value changes. The weighted average, average similarity score, and extreme value change regularity are combined to generate a periodic regularity coefficient.

7. The multi-sensor data based real-time pose regulation method for a tunnel lining jumbo according to claim 1, characterized in that, The methods for calculating the concentration and dispersion indices of the characteristic components include: Singular value decomposition is performed on the spatiotemporal feature matrix, and the singular values ​​are arranged in descending order to form a singular value spectrum; An adaptive thresholding algorithm is used to divide the singular value spectrum into primary component regions and secondary component regions; Calculate the cumulative contribution rate of singular values ​​in the principal component regions and use it as a concentration index. The degree of difference between the singular value distributions of the primary and secondary component regions is analyzed, and the degree of difference is used as a dispersion index.

8. The multi-sensor data based real-time pose regulation method for a tunnel lining jumbo according to claim 1, characterized in that, The method for calculating the quality assessment coefficient includes: Calculate the correlation matrix between the row vectors of the spatiotemporal feature matrix and analyze the internal structural features of the matrix; The internal structural features are coupled with the periodicity coefficient to obtain the regularity score of the sequence. The concentration and dispersion indices of the feature components are combined and transformed to obtain the feature integrity score of the sequence. The quality assessment coefficient is generated by combining the regularity score and the feature integrity score through weighted fusion.

9. The multi-sensor data based real-time pose regulation method for a tunnel lining jumbo according to claim 1, characterized in that, Methods for dynamically adjusting pose data processing strategies based on quality assessment coefficients include: The intensity parameters of pose data filtering are determined based on the quality assessment coefficients, and a mapping relationship between the filtering intensity and the quality assessment coefficients is established. Based on the filtered pose data, the real-time pose state of the trolley is calculated. Based on the deviation between the real-time pose state and the desired pose, a pose adjustment command is generated. The posture control of the tunnel lining trolley is achieved by executing posture adjustment commands through a closed-loop control system. 10.A system for real-time regulation of a tunnel lining jumbo pose based on multi-sensor data, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the real-time position and posture control method for tunnel lining trolley based on multi-sensor data as described in any one of claims 1 to 9.

Images